Changing from System to Ecosystem
When IBM first presented its Watson system on the game show Jeopardy! in 2011, it was like something out of science fiction. Here was a computer that could not only understand spoken questions, but answer them quicker and more accurately than the best human players. Nobody had ever seen anything even remotely like this before.
Today, less than a decade later, artificial intelligence has been transformed from the amazing to something approaching the ordinary. Not only do we have capabilities that are similar to the original Watson system on our phones, we can access top-notch resources from multiple companies, often for free.
Yet perhaps the largest difference from those early days of Watson is how AI has evolved from a system to an ecosystem. Much like the evolution from mainframes to personal computers, companies now can access various components to build a system created specifically for their business. That’s how a technology becomes transformational.
Acquiring Data Resources
What makes artificial intelligence different than earlier technologies is that the system learns as data is put into it. So data is the fuel that makes the whole mechanism function. That’s why Josh Sutton, CEO of Agorai, a platform that helps companies build AI applications for their business, calls data, “most valuable business asset that does not reside on a balance sheet.”
“The first thing I suggest a company asks is ‘what value can I create with AI?’” Sutton told me. “From there, you can figure out what data you need to drive those applications, what data assets you already have, what you can generate and what you need to buy. I would also look at what data I have that may be valuable to others, but not of strategic value to me, that I can use to generate a revenue stream.“
Besides, he emphasises that because AI algorithms can adapt the way the application works to the data you feed it, these technologies are much more accessible to small and medium sized businesses who can’t afford expensive consultants to customize a whole system for them.
David Schatsky, a Managing Director at Deloitte who concentrates on emerging technology trends, points out that advances in synthetic data techniques are making it possible to train a system even in industries where actual data is difficult to obtain. That means that very high performance can be obtained for a reasonable cost even in highly regulated fields.
“In healthcare, for example, where much of the data is protected, it’s very difficult to access the data we need to create a good model using machine learning. However, new techniques like synthetic data generation allow us sidestep that problem so we can build systems that automate processes and improve performance” he said to me.
Advancing Both Software And Hardware
In the Master Algorithm, the renowned machine learning researcher Pedro Domingos says that there are five basic machine learning approaches, including neural networks, vector machines, genetic algorithms and inductive learning algorithms. IBM’s Watson, in fact, included a number of these approaches to beat human contestants on Jeopardy!
Since 2011, in part because of the excitement that Watson created, these approaches have advanced dramatically. For instance, Google’s Brain project was able to significantly raise the amount of layers in a neural network system, which made its algorithms much more accurate.
Another significant field of advancement is in hardware. The chipmaker Nvidia’s expertise in graphics processors has helped it dominate the market for AI chips. Google and Microsoft are building their own chips specifically designed to run their algorithms. IBM developed neuromorphic chips which mimic the design of the human brain which, because they use very little power, have the potential to bring AI capabilities to edge computing.
Most business never see these technologies directly, but their advancement is what drives the capabilities of AI applications. It is one level up the stack that business value starts to be built.
While algorithms and hardware are essential, what makes applications beneficial is their ability to mimic human intelligence. Things like machine vision, language analysis, document summarization and process automation are the basic building blocks of any intelligent system.
Typically, these can be accessed for free through open source platforms, such as Shogun, or through the cloud computing platforms of the major tech firms. For instance, Amazon, Google and IBM all make API’s available to developers. Often these are used as a loss leaders for their cloud platforms, so are free to anybody who wants to use them.
Working with these capabilities directly takes a lot of expertise. Nevertheless, even amateurs with little or no technical skills can access them through no-code platforms. So a marketing manager who wants to create a simple chatbot can access API’s for natural language processing and text-to-speech capability without writing a single line of code.
Where business value really starts to be created is when these capabilities are bundled into solutions and then trained on data for a certain task. Increasingly, for tasks that are fairly common, there are pre-trained models that can plug into existing systems, with relatively little need for expensive integration.
“What we’re seeing with our clients is that there a lot of well defined problems where there is no need to reinvent the wheel,” Schatsky of Deloitte told me. “So pre-trained models can be enormously helpful in automating basic processes. That frees up resources that can be focused on more specialized systems and can help create a true competitive advantage.”
Creating Business Solutions
What’s most essential for business leaders to know is that AI is no more some type of “gee whiz” technology, but increasingly key to competing effectively in today’s marketplace. As the technology keeps evolving from difficult integrated systems to a modular ecosystem, even small and medium enterprises will find that they need to adopt these capabilities or fall behind.
“What most business leaders aren’t always aware of, but are quickly waking up to, is that there are an untold number of industry and process specific AI applications that can accelerate their business,” Sutton says. He speaks about companies like Primer.ai, which summarizes massive data sets into analyst reports for financial companies and Daisy Intelligence, which optimizes supply chains for retail companies, as just two examples.
“At Agorai, we’re focused on helping to connect those businesses to the companies offering solutions that can help them,” he told me. “What I tell business leaders is that AI is useful for tasks you understand well enough that you could do them if you had enough people and enough time, but not so useful if you couldn’t do it with more people and more time. It’s a force multiplier, not a magic box.”
Schatsky of Deloitte sees many of the same trends. “There is a relentless pace of progress, the tools are improving, the models are improving and the systems are improving and we’re seeing companies achieve real results. It’s still early days, but the future is really exciting,” he says.